Optical Coherence Tomography-based Factors Associated With Visual Loss in Diabetic Patients

The aim of this study is to evaluate the relationship between retinal structures and visual acuity in diabetic patients using optical coherence tomography (OCT) and OCT angiography (OCTA). This study was a retrospective observational study conducted at a single medical center in Japan. Evaluation of retinal images was analyzed using spectral domain OCT. Twelve factors including central retinal thickness, length of disorganization of retinal inner layer (DRIL), number of inner hyperreective foci, number of outer hyperreective foci, height of intraretinal uid, height of subretinal uid, length of external limiting membrane disruption, length of external ellipsoid zone (EZ) disruption, vessel density of supercial capillary plexus (SCP), foveal avascular zone (FAZ) area, and FAZ circularity were analyzed based on OCT/OCTA ndings. Multivariate analysis was used to investigate the OCT-based factors that could be correlated with poor visual acuity in treatment-naïve diabetic eyes. A total of 183 eyes of 123 diabetic patients with type 2 diabetes (mean age 61.9 ± 12.3 years, 67 men and 57 women) and 62 eyes of 55 control subjects (mean age 64.4± 12.5 years, 15 men and 40 women) was enrolled in this study. Multiple regression analysis showed that OCT-based factors correlated with visual acuity were length of DRIL (β = 0.24 (cid:0) P < 0.01), length of EZ disruption (β = 0.35 (cid:0) P < 0.001), and FAZ circularity (β = -0.14 (cid:0) P < 0.05). The other factors showed no signicant correlation. In conclusion, the length of DRIL, length of EZ disruption, and FAZ circularity measured by OCT were identied as related factors for visual impairment in treatment-naïve diabetic eyes. controls used the Mann-Whitney U test. Comparison of clinical factors, OCT/OCTA-based factors at various DR stages used the Kruskal-Wallis and Steel-Dwass tests. Multivariate analysis was performed to identify factors signicantly associated with visual acuity. Statistical analysis was performed using commercially available statistical software (SPSS version 21.0). P values < 0.05 were considered to indicate statistical signicance.


Introduction
The global prevalence of diabetes mellitus (DM) is estimated to have affected 415 million people in 2015, and is expected to involve 624 million by 2040 [1]. Diabetic retinopathy (DR) is the most common ocular complication, which is asymptomatic at an early stage, but progression of the disease leads to severe vision loss [2]. DR is the leading cause of blindness in working-age adults worldwide and is estimated to affect onethird population with diabetes [3]. Diagnosis of DR relies on the detection of microvascular lesions, which is divided into two stages based on their clinical ndings: nonproliferative diabetic retinopathy (NPDR) and PDR [4]. The main vision-threatening complication of DR is diabetic maculopathy, including diabetic macular edema (DME) [5] and diabetic macular ischemia (DMI) [6], and the presence of vitreous hemorrhage, traction retinal detachment [7] or chronic in ammation within the proliferative membrane [8] in PDR. Although treatment of DR remains challenging, timely interventions in eyes at a risk for DR progression have been mandatory for preventing visual impairment [9].
Optical coherence tomography (OCT) is a non-invasive and fast imaging modality that offers imaging of crosssectional structures of the retina using low coherence interferometry and captures high-resolution twodimensional images from scattered light from different layers of the retina [10]. This technique has contributed to the detection and monitoring of DME [11] and DMI [12]. Several studies have shown that con gurations of ellipsoid zone (EZ) are associated with visual function following treatments for DME [13][14][15]. OCT angiography (OCTA) is an evolutional development of conventional OCT that detects movements of red blood cells and blood ow contrast, visualizing retinal and choroidal microvasculature without the need for dye injection [16].
In fact, repeated scans at the same location can detect the changes in OCT re ectance signals from the ow through the blood vessels [17]. Recent studies analyzing the role of OCTA in DR have illustrated some quantitative OCTA metrics as linked to the severity of DR and DME [18]. Among the quantitative metrics, vascular density [19], vascular diameter index [20], total length of vessels [21], vascular architecture and branching [22], and area of the foveal avascular zone (FAZ) [23] were reliable OCTA-based vascular factors. Thus, OCT and OCTA may help quantify various retinal pathologies such as neurodegeneration, retinal edema, and capillary dropout (reduced vessel density) in DR, thereby assessing their relevance to visual function. However, it has not been fully elucidated which related factors based on OCT(A) ndings are well correlated with visual acuity in treatment-naïve DM patients. The purpose of this study is to evaluate the relationship between various retinal structures and visual acuity in DM patients at various stages of DR using OCT and OCTA.

Results
Clinical characteristics of control and diabetic patients A total of 183 eyes of 123 diabetic patients with type 2 DM (mean age 61.9 ± 12.3 years, 67 men and 57 women) and 62 eyes of 55 control subjects (mean age 64.4± 12.5 years, 15 men and 40 women) satisfying the inclusion criteria was enrolled in this study. Table 1 shows the clinical characteristics of the subject patients.
There were no signi cant differences in age, intraocular pressure, or axial length between the control and DM groups. On the other hand, BCVA was signi cantly lower in the DM group compared with the control group (P < 0.05). As shown in Table 1, the frequencies of hypertension and dyslipidemia were signi cantly higher in DM group than in control group (P < 0.05 and P < 0.001, respectively). The DM group (n = 183) was classi ed into three groups, NDR (n = 92), NPDR (n = 67), and PDR (n = 24). There were no signi cant differences in age, intraocular pressure, axial length, duration of diabetes, frequency of hypertension and dyslipidemia among the DM groups. In contrast, the more severe diabetic retinopathy had signi cantly lower visual acuity (P < 0.001) and higher HbA1c P < 0.005). First, this study examined the difference in OCT-based factors between control eyes and all diabetic eyes ( Table  2). There was no statistically signi cant difference in ELM, VD of SCP, and FAZ area when comparing control eyes and all diabetic eyes. However, CRT, length of DRIL, number of HF, height of IRF, height of SRF, length of ELM disruption were signi cantly larger (each P < 0.05) and FAZ circularity was signi cantly smaller (P < 0.001) in all diabetic eyes than in control eyes. Next, when all diabetic eyes are classi ed according to the stage of DR and compared, CRT, length of DRIL, number of HF, height of IRF, height of SRF, length of ELM disruption, length of EZ length, VD of SCP, and FAZ area were signi cantly large (each P < 0.05), and FAZ circularity was signi cantly small (P < 0.001) as the severity of disease increased.

Correlation between OCT-based factors and visual acuity in diabetic eyes
In this study, multivariate analysis (multiple regression analysis) was performed to evaluate the effects of various OCT-based factors on visual acuity in diabetic eyes (Table 3). Multivariate analysis showed that length of DRIL (β = 0.24, P < 0.01), length of EZ disruption (β = 0.35, P < 0.001) and FAZ circularity (β = -0.14, P < 0.05) were signi cant factors affecting visual acuity. There was no signi cant association between the other factors and visual acuity. As a result of multivariate analyses, this study further examined the correlation between each factor of OCT images related to visual acuity in diabetic eyes. (Table 4). There was a signi cant positive correlation between length of DRIL and length of EZ disruption (r = 0.71, P < 0.01). FAZ circularity had a signi cant negative correlation with length of DRIL and length of EZ disruption (r = -0.33 and -0.31, both P < 0.01). worse visual acuity in DME, and the changes in DRIL predicted subsequent visual outcome [26,27]. Ocular disorders caused by retinal vascular dysfunction are likely to complicate disruption of the blood-retinal barrier (BRB), followed by extracellular uid accumulation in the intraretinal or subretinal space [28]. Although the mechanism of DRIL formation is unknown, it is hypothesized that DM-induced microvascular damages inside the retina may represent structural deformation identi ed as DRIL based on OCT images [29]. In addition, the PDR may form a brovascular membrane that creates traction on the macula. Therefore, multiple mechanisms associated with vascular abnormality and mechanical stress may be involved in DRIL formation. This can lead to deformation or breakage of the synaptic junction between photoreceptors and retinal ganglion cells.

Tables 3. Multiple regression analysis of the correlation between OCT-based factors and visual acuity
Furthermore, although a histological observation of DRIL has not been demonstrated, DRIL re ects damage of various cells such as Müller cells, bipolar cells, horizontal cells, and amacrine cells, and is believed to affect the visual outcome of diabetic eyes [27]. Our results, showing the correlation with the length of DRIL and visual acuity, support the ndings of previous reports, and the widespread presence of DRIL may contribute to visual dysfunction in treatment-naïve diabetic eyes.

1-2. EZ disruption
It is indisputable that the retinal photoreceptor layer can be accurately evaluated by examining the integrity of the EZ using OCT, which corresponds to the inner segment of photoreceptors, indicating that EZ integrity has been shown to be strongly correlated with visual acuity [30]. Although the mechanism of photoreceptor degeneration is not yet well de ned, several processes may lead to photoreceptor damage. Destructed BRB associated with DM may induce extravasation of blood components and in ammatory cell in ltration, causing morphological abnormalities in the macula [31]. Uji et al. reported that HF is considered a lipoprotein, a precursor of hard exudates, which was incorporated by photoreceptors or macrophages, leading to poor visual acuity [32].
Murakami et al. suggested that incomplete EZ often exists under the cyst space, and that the cyst space of the outer plexiform layer may contribute to photoreceptor damage in DME [33]. In our study, EZ disruption was found in 10.9% of all DM patients, 10.4% in NPDR and 54.2% in PDR, which occurred more frequently in the advanced stages of the disease. Our study also con rmed that the patients showing high frequency of EZ disruption had signi cantly poorer visual acuity.

1-3. FAZ circularity
DR is primarily a disease of the retinal vasculature, and OCTA provides valuable information about vascular conditions not available with structural OCT. FAZ is a specialized area of the retina containing the highest density of cone photoreceptors and high oxygen consumption [34]. As the FAZ of the healthy eye usually exhibits a circular or elliptical shape, deviations from this shape are common in vascular pathology [35,36].
Balaratnasingam and colleagues used OCTA to reveal that FAZ area correlates with visual acuity in patients with DR, and proposed FAZ as a biomarker of visual function [29]. FAZ circularity is de ned as the ratio of the FAZ boundary to the perimeter of a circle of equal area [35]. Krawitz and colleagues found an association between the disrupted circularity and DR severity, but reported no correlation with visual acuity [37]. On the other hand, Tang et al. demonstrated that FAZ area was not a sensitive marker that correlates with central vision, while decreased FAZ circularity is associated with impaired visual function [38]. In our study, we showed a correlation between visual acuity and FAZ circularity, in which eyes with poor visual acuity had a more irregular shape of FAZ. Alipour et al. explained that studies using the digital curvelet transform revealed that the FAZ of DR deviated from the mildly wavy boundaries found in healthy controls, which became apparent in severely diseased eyes [39]. Defects in the FAZ margin are mainly due to capillary loss and vascular remodeling. The FAZ circularity index allows quanti cation of disruption of the terminal capillaries in the fovea, which may be a more relevant measure of visual acuity. In addition, macular edema can result in decreased vascular elasticity due to mechanical stretching that results in blockage of blood vessels [40], and can also change FAZ contours [41].
Therefore, assessing both the size and shape of FAZ is probably important for detecting pathological alteration of the macula in DR.

Changes in outer and inner retina
In summary, our study showed that the following 3 OCT-based factors could play important roles in impaired visual function in diabetic eyes: length of DRIL, EZ disruption, and FAZ circularity. In addition, these factors had a signi cant correlation with each other in treatment-naïve diabetic eyes. However, there is a lack of information on how these factors correlate with each other. This study further discusses three categories: 1) DRIL and FAZ circularity, 2) DRIL and EZ disruption, and 3) EZ disruption and FAZ circularity.

2-1. DRIL and FAZ circularity
Recent studies suggest that DM eye degeneration can be caused by two different states: vasculopathy and neuropathy [42]. NDR and mild NPDR have demonstrated thinning of RNFL due to changes in microcirculation [43]. The metabolic and oxidative stresses of diabetes lead to increased sensitivity of ganglion cells, leading to neuronal loss [44]. After that, DRIL occurs with the severity of retinopathy, which may be negatively correlated with FAZ circularity in SPC. In fact, the frequency of DRIL and FAZ circularity found in this study was signi cantly high as DR severity increased. Taken together, the relationship between DRIL and FAZ circularity could be associated with both neurovascular damages during DR progression, although segmentation errors in OCTA might exhibit.

2-2. DRIL and EZ disruption
The sub-analysis of this study showed correlation with DRIL and EZ disruption, suggesting that the extent of DRIL may also be involved in the retinal outer layer in DR. It remains unknown how different cell types die during DR progression. DR is associated with loss of pericytes and endothelial cells in the retinal vasculature, leading to loss of the capillary and BRB [45]. The deep capillary plexus (DCP) including the INL and the OPL are probably important for nutrition to photoreceptor synapses, as the retinal circulation is responsible for approximately 15% of the oxygen supply to photoreceptor inner segments in the darkness [46]. Therefore, photoreceptor cells in the presence of DRIL may be susceptible to ischemic insults in DCP. The presence of DRIL indicates an anatomical disruption of visual transduction pathways in the medial retina and may interfere with the transmission of visual information from photoreceptors to ganglion cells. The anatomical changes observed in this OCT may cause the association with DRIL and EZ disruption.

2-3. EZ disruption and FAZ circularity
It has also been shown that diabetes is associated with increased loss of retinal neurons [47]. Retinal glial cells including astrocytes, Müller cells and microglia are involved in structural support and maintenance of homeostasis in the retina [48]. Under hyperglycemic stress, microglia are activated, various cytokines are secreted, and astrocytes and Müller cells are involved with the ampli cation of the in ammatory response by the production of proin ammatory cytokines [49]. Previous studies have reported that FAZ measurements of both SCP and DPC correlate with EZ and IZ disruption, and photoreceptor morphology and integrity are susceptible to macular ischemia [50]. Long-term or more severe macular ischemia not only contributes to the onset of macular edema, but may also cause retinal atrophy due to photoreceptor damage.
It remains open to debate whether diabetic retinal neuropathy is the cause or effect of microangiopathy under chronic hyperglycemia. Barber proposed two hypotheses to explain the relationship between neurodegenerative and microangiopathy changes [51].

Limitations
We are aware of some limitations of this study. First, this report includes a retrospective design and a limited number of subjects. A larger sample size in prospective studies is important to recon rm the results. Second, it was impossible to investigate the VD of the DCP, as the OCTA parameters automatically calculated in AngioPlex relate only to SCP. However, this technique produces projection artifacts in most cases, so SCP analysis is more accurate than DCP one [52]. Third, if the capillary ow rate detected by the OCTA algorithm is below a threshold, the terminal capillaries may not be fully visible. Therefore, measurements of FAZ area and VD may be affected in selected cases. Fourth, demographics such as age, duration of diabetes, HbA1c levels, and hypertension were not used in the regression analysis. This needs to ensure a larger sample size and will reveal a correlation with vision loss. Fifth, the biomorphometry performed in this study utilized data from a single visit. Therefore, it is impossible to use this result to judge the risk of disease progression or to predict response to treatment for DR. Further studies are needed on the natural history of DR and the correlation between post-treatment OCT-based factors and visual acuity. Sixth, we could not examine the degree of pathological alteration and visual impairment in the outer and inner retinas con rmed by OCT images. We need to secure more samples in the future to investigate how parafoveal signaling is related to the degree of visual impairment.

Conclusions
The length of DRIL, length of EZ disruption, and FAZ circularity measured by OCT were identi ed as candidate factors for visual impairment in treatment-naïve diabetic eyes. OCT and OCTA image acquisition

Patients
The AngioPlex OCTA software featured an OCT microangiography algorithm that incorporated differences in both phase and intensity information, and a FastTrac retinal tracking system to reduce motion artifacts [24]. OCTA parameters OCTA images of the SCP were generated using an automated software algorithm ( gure 3A, D). The upper boundary of the SCP slab is the ILM and the lower boundary is an approximation of the IPL inferred by the following equation (1): where Z, T, and OPL indicate the boundary location, thickness, and outer plexiform layer, respectively. Several quantitative parameters were evaluated from OCTA images automatically generated from Angiography scan 6 × 6 mm slabs according to previous reports: vessel density (VD), foveal avascular zone (FAZ) area, FAZ circularity [25]. VD was de ned as the total area of perfused vasculature per unit area in the measurement area, determined by summing the number of pixels with ow signal and dividing the total number by the number of pixels in the study area. Quantitative assessment of VD was obtained from the central (